Transcription
So, we've reached our final session of the day. We're closing out with a look at how DBS, Southeast Asia's largest and most profitable bank, is betting big on AI. Building systems that are fast and fair while staying ahead in one of the world's most competitive markets is a seemingly neverending challenge. Making AI work at this scale requires more than just tech; it requires leadership, trust, and a clear vision for the future. Please join me in welcoming a leader at the forefront of that shift in finance, DBS Group CEO Tan Suan. She'll be speaking with Fortune's Clay Chandler. Thanks very much.
[Music]
>> Thanks, Andrew.
>> Susan, please.
>> Thank you.
>> I cannot tell you what a pleasure it is to have you here with us this afternoon. uh uh Susan returns from uh from uh last year when she also spoke uh at uh at Brainstorm AI. She has since been given a new role and responsibility. Congratulations.
>> Thank you.
>> Uh the CEO of uh of Singapore's biggest bank, the most profitable uh company on our global uh Southeast Asia 500 list uh for the second year in a row. And I would say also that uh Susan, we just put out our most powerful our global most powerful women list. Susan was number six. Wow.
>> The only uh woman from Singapore uh on to make the top 10. So congratulations on that front. uh as well. We were very excited to have uh Susan uh uh join us uh here today because DBS is well known for being so actively engaged in thinking through analytics uh, you know, the digital revolution and uh especially more recently how to use artificial intelligence in the operations of the bank. They're thinking through every angle of this new technology and what it means. They are thinking about what it means for customers, how they can serve customers better. They're also thinking about how they can use it internally uh at the bank and uh very exciting right on the kind of cutting edge here uh in Singapore in in leveraging these new technologies.
Now, Susan, I want to I want to ask you about how much uh of your time thinking about AI is consuming, but maybe you could start by telling you the kind of you your board kind of gave you notice that you were going to have to focus on AI when when you came into the job. Explain that.
>> The day the day my u my CEO role was announced, I get I I'm in a WhatsApp group with my board. I get a the WhatsApp saying, "Even the CEO's job, CEO's job will be replaced or can be replaced by AI." So therefore, it's the tone from the top, right? So I'm like, okay, if I can be replaced by an AI, so can everything else. Agree. And that's why my New Year message to my team was was my four Rs. We've got to reinvent ourselves. We've got to really stay relevant. Um, we've got to be resilient because it's going to be a volatile and turbulent ride. Um, and we have to be responsible, right? So that's my four Rs. I think the four Rs still stay highly relevant um today and and and probably in time to come. Um, and I tell my managers, don't hire for knowledge because knowledge is ubiquitous. Knowledge has been democratized. Hire for attitude because you want to hire people who are agile, uh, who are humble, who are able to say, "Whatever I knew up to today is no longer relevant today or tomorrow. Therefore, I have to not be afraid to let go, take a step back, relearn some new skills to go forward."
>> Are people receptive to that or does it uh invoke a lot of uh you know uh
>> So, you know, when we first rolled out the generative AI experiments and and we have both horizontal and vertical use cases, right?
>> You know, what was interesting?
>> The most junior people were using it. Fair enough, because you get a lot a lot more to learn if you're entry-level staff. And the most senior people were using it because they knew that, you know, in a way, it It was us trying to download their brains into the system. So they were testing it and they were using it as well. It was the middle that was the least engaged. But I think it's that middle that is most at risk if you don't engage. Right. So today now, the the the the trend has turned because I think the most junior and most senior people have made it safe for everyone in the middle to also start using it. So, and and we use this for so many different things. We've got a horizontal DBS GPT where you can use it for "ask me anything" to vertical use cases where, you know, we come up with agents that are very specific. You know, it can be for low-level transactions, it could be for higher-level, higher-order reasoning, making of pitch books, writing credit memos, to answering ESG questions, for example, or wealth co-pilot to the horizontal, you know, inch deep, mile wide, where we can say, "Ask me anything." It could be a branch manager saying, "How do I get this KYC form filled?" to "Who do I call for, you know, helping with an industry query?" to "Translate this email for me, please. My my English or Chinese isn't that good." To whatever. So, you know, compliance, filling out forms, writing emails. I I that makes perfect sense. I think that's something all of us are are are using this new technology for. But what's amazing to me is that you're looking for ways to to use the technology even at some of the most uh high-level and sophisticated transactions that the bank is involved with. I mean, can you give us some examples of kind of significant transactions where AI has come in useful?
>> I think it's, you know, to me, it's deterministic and probabilistic.
>> Yeah.
>> For stuff that's probabilistic where you have to somewhat predict the future, I think it's going to be tough.
>> Right.
>> So, for example, if you ask an AI, you know, "Tell me where the dollar yen is going." Uh, and you know, we started AI for wealth management in 2014 with IBM Watson and we realized there was time, and then there was also recency latency, and and and what is your horizon, what's your risk profile, etc., and the answer may not be the right answer for for the right person. But if it's deterministic, like, "What is the best card for for air miles?" or "How do I optimize my B?" Again, optimizing of balance sheet may not be deterministic, but at least you can come up with option one, two, and three for the customer to choose. Or, you know, um, "What's the the best mortgage in the market today for for Sing dollars if I want a floating package or a fixed rate package?" et cetera, et cetera. Um, and frankly, service calls, right? Today, a lot of our service calls um can be managed by a generative AI bot, right?
>> Uh, for me, we rolled it out uh first to 10,000 people, then to 50,000, then to 100,000. And the fact is, once it works, I think once it crosses 90, 95% accuracy, um, we're good to go. And if it's not, if it makes a mistake, we will get, we will say, "Call a human now," right?
>> So you must always have a human in the loop just to make sure
>> that it doesn't hallucinate.
>> But we also test the the the LLMs one against the other so that we can also play them off each other.
>> That's really important because as anyone who's who've used these uh platforms knows, they can be very sycophantic. They can say, "Oh, that's a brilliant question. It was so smart of you to ask that." And when you kind of give them prompts, they can lavish praise on your prompts. And with a bad prompt, you can come up with an answer that's just kind of self-indulgent. Uh, but it so you've got what uh I don't know, ChatGPT giving an answer and then Proximity or whatever coming in and saying, "Oh, ChatGPT, that's rubbish."
>> So, we we we we often use two different LLMs to to compete before we give the customer the final answer, right? And that's because we want to make sure that they're not so different from each other that you might have a risk of hallucination or just the wrong, just the wrong outcomes.
>> Right.
>> Yeah.
>> But as I said, human in the loop constantly now, especially when it's so new, human in the loop is important,
>> right?
>> But also, you know, for for a lot of companies who who are going through this journey, the first thing is get your hygiene right. And what is your hygiene? Your hygiene is making sure your data is stored correctly, right? That you have your data in a data lake, your metadata in a data lake, and then you have your models and your protocols set up.
>> Right.
>> Right. So how do you set up your models and how do you? Because in the past, we had our models in different parts of the bank, but now pulling it all together, having one common protocol, having a common language, um, and then having a common philosophy around how you guard it, and now also increasingly having the end-to-end data governance is important. When your data is created to where it's purged, >> how and who is in charge along the way uh of that part, which part of the data is important.
>> Um, so there's a lot to think about, but start, it all begins and ends with the data.
>> Yeah.
>> And how you you you store it, how then you create models, and when the models work or don't work, how do you have that feedback loop? Then when you bring in LLMs, how do you make sure that your customer data, if you're using the compute of the LLMs, how it's purged as well, so it doesn't go back to the LLM.
>> Um, and then having that constant feedback loop >> both from your customers and from your employees. That's a really great roadmap for lots of companies because, you know, certainly what we hear at Fortune is that companies are struggling to get beyond the kind of proof of concept where they sort of have a little siloed experimental use uh of AI to see what happens to then moving to this more siloed approach that you describe where, okay, uh, you know, each of my different business units or verticals are experimenting with AI in a different way. But what you're describing is the kind of holy grail of this stuff, which is a much more holistic, uh, bank-wide uh, corporation using it not just for customers, but for uh, for employees as well. I'm curious, organizationally, what does that mean for how you structure the bank? I mean, I spent a fair amount of time with the folks at uh DBS uh last year, and I remember being very impressed by how large the data analytics and and kind of, you know, uh, digital team uh is. But you've restructured the uh, the whole organization to make sure that it's much more kind of uh, unified in the way that you think about AI.
>> What I did um as the new CEO was I created a COO function, a Chief Operating Office, but really that's like a Chief of Staff, someone who, some, a big group that oversees the end-to-end.
>> Right.
>> They must be product, segment, and business agnostic,
>> right?
>> But they make sure that end-to-end we're good. So my data and transformation people sit there.
>> Yeah.
>> Um, because as we transform the bank, it's end-to-end. It's not just product, customer-facing. It's also operations, it's technology, it's data. The transformation must be holistic and end-to-end.
>> Um, and it's also legal and controls and governance, right? Because, right, >> your data security must must must work. Um, as a regulated entity like a big bank, you need to abide by all your different regulators and the different regimes you sit in. So, you're making sure the governance framework works. Um, making sure the customer experience is good, the pricing is correct. So whilst business is very much on that front, it's also learnings from different parts of business. So retail and consumer might do something with their chatbot, and then corporate might do something with their chatbot, but actually, they should learn from each other, right, and optimize and and see how the generative AI can help our chatbots be stronger and better and more intuitive in the answers from the first go, from the get-go. And we've always said first call resolution is important. So both the consumer and the corporate bank now are focused on getting that first call resolved, hopefully entirely by a bot, right? Then you release a lot of capacity. Now, what do we do with this capacity? So HR comes in and um, today, a lot of um HR training is done by a bot. Um, yesterday, Kha, our minister, was at our launch when we launched executive coaching done by um, a digitized version of Marshall Goldsmith. Now, Marshall Goldsmith is a world-renowned executive coach, and not everyone can afford him, but an AI or generative AI version of him can be done for for all and sundry. And that's exactly what we've done. We've we've we've digitized the generative AI of an executive coach and we've enabled everyone, every one of our staff now has this coach in the palm of his or her hand to say, "Listen, my job has been taken over by AI. What do you think I can give me some suggestions on what I can do for my next gig, right? Or my next role, and what what do I need to learn to to get into that role? I'm interested in wealth management. I'm interested in in in in me banking, whatever it is, help me through this journey." Um, and and it does that. It gives you answers. It helps with, it's, you know, active, your active personalized AI coach, right? Executive coach, which I think is great. We launched it yesterday.
>> Just yesterday. So, this is important for you though, because you're basically telling people you should embrace the turbulence and uncertainty.
>> Don't be scared of it. Face it head on.
>> Yeah.
>> Face it head on. And I think Singapore's all in on AI, generative AI. Um, and and DBS is all in. uh we see profound changes coming um, and we're excited by the changes, and I think um, it will make us so much more productive and and and effective. Uh, in fact, in my in our town hall that we did uh for the leaders, we said this is going to make you a superhuman banker or a superhero banker because you can literally do everything really quickly. Um, I remember going to a client pitch quite unprepared and um, you know, talking, chatting as as the client was talking to me, I had my phone under the table. I was like, "Oops, the deal has changed, and now they want to do this instead of that. Help me think of, help me think of questions to ask him." So that came up, four questions to ask the client, and I kind of read off and I said, "Have you thought about this for for your business or that for your business?" And the client said, "Wow, Susan, you know my business so well." So, I cheated.
>> But is that cheating?
>> I think we can all cheat now.
>> Is that cheating? I mean, it's a good question, right? I mean, the tools are there.
>> The tools are there, and you're dumb not to use it, right? So,
>> um, you know, we tell everyone, use it, but it doesn't replace the human interaction and the human um, conversation. So, the client still wants to have that face-to-face, that eye-to-eye.
>> Yeah.
>> Um, and you want to read the the body language of the promoters or their family members or their CFO, CEO, chairman, etc. You still need a human to bring the deal across the line. Yeah.
>> But a lot of the grunt work, the data crunching, the pitchbook making, the capital markets optimization, etc., all that can be automated and done by a bot. Then you just go in and you try and seal the deal or you try and get the deal across the line. So this is fascinating to me, and and we've had a lot of discussions uh over the last two days about the idea of agentic AI, moving from AI that has to be get specific instructions and prompts for every little thing to AI that can exercise some form of judgment that it can um, and I was watching uh I was watching a speech by uh Yuval Noah Harari the other day, and he was making the point that, you know, financial services is the cutting edge of this agentic stuff because everything is information-based. It's all, there's all data. So it's um, you know, very uh susceptible to being completely overtaken by the technology. And he raised the question, >> how willing are we to let this agentic stuff go in financial services? Are we willing to say, "Hey, an AI can open a bank account?" Hey, uh, we're going to let uh the AI decide how to invest the assets in that bank account. Does it go in the stock market? What does the portfolio look like? Um, does it uh invest in a new business? Yeah. You know, that sort of thing. Are you thinking through those kinds of questions? Because customers may come to you and want them.
>> I think with, we're thinking about it, as I said, end-to-end. First of all, it's the point of entry, right? Does the customer look for you through their own agent, right?
>> Do they ask their personalized agent, "Hey, find me the best bank for me banking, find me the best bank for wealth management, find me the best bank for da da da da da." Right? So the point of entry could be agent-based.
>> I see.
>> And if that's the case, how are you positioning yourself? In the past, we use, in the past, even now, we use search optimization, SEO, etc. In the future, do you move that >> to make sure that you are positioned in the agentic space so that the that all your potential prospects or clients' LLMs or agents can still get to you? um, or will there be then, you know, paid advertising again in the LLM space? Who knows, but that's happening as we speak. So I've told my team, be prepared for a world where the customers come through us through their agents, right? And then it's agent to agent communicating with each other, right, right, and making sure that we're at the forefront of this.
>> So that's that's happening um, and then take, and then also you've got the whole new digital asset ecosystem, which is not today's topic, but that's also happening. The world of tokenization and, you know, uh, smart contracts and stablecoins, etc. That whole ecosystem is also taking place. So there's a lot of very exciting big changes happening in in in in in our world, especially in the financial industry. And as I said, you have to embrace it to stay ahead. But you also have to have a mindset where, you know, we're pioneers, we're all new to this, so we're going to have to do some experiments and sandboxes. We're going to have to tolerate some some errors and some mistakes, learn from that. So the key thing is, I told my staff, you know, when you try something, please share your experience, what worked, what didn't work, especially what didn't work,
>> right?
>> So that you can save some other people time,
>> right,
>> and effort, u and mistakes. Um, but but but really, this fact that you brainstorm together, I we have brainstorming, a lot of brainstorming together as as a leadership team um, and and then our job as a leadership team is to prioritize what to roll out first,
>> right?
>> Prioritize what's important, what moves the needle, uh, what keeps us ahead, um, and then saying no to some other things that may not, may not move the needle too much.
>> I mean, if you were to sort of single out the one or two biggest areas where you see the capacity to really move the needle, what would they, what would they be for you?
>> So firstly, it's really on on on service and operations that can really move them, and it already is. As I said, you know, in in terms of side, 90 over% of our service calls can be done by generative AI. Think about how much that releases a lot of capacity, right?
>> And and now I can move a lot of our service staff into becoming relationship managers,
>> because I have so many, they don't have relationship managers now. We can actually thank them.
>> Right.
>> They can actually own customers and look after them human to human. Really bring that level of relationship up from, "Can I check if my payment's been done?" that a bot can do, to, "Hey, how are you thinking about supply chains in an era of tariffs? Right? How can we help?"
>> Right, right.
>> So you, you, you, you're really going higher level, so to speak.
>> Um, so I'm excited about the the the the changes in people's careers. That's why we said, you know, even AI for career coaching is important. Um, and I'm excited about how we can reshape the workforce for the future.
>> This has just been a tour de force. So Jean, uh, I cannot think of a better speaker to kind of wrap up our our two-day event here. We've all been brainstorming just in the way that you've been brainstorming uh at the bank. Very inspiring to see the kind of enthusiastic, forceful way that you are trying to embrace this uh new technology at DBS, and we will will watch your progress with great interest. Please thank Susant for joining us.
>> Thank you. Thank you.
>> Thank you.
>> Well, that was a tour de force. Uh, really appreciate uh Susan's remarks. That brings us to the end of Fortune's Brainstorm AI Singapore for 2025. Thank you for staying with us right till the end here. On behalf of Fortune, I want to extend a heartfelt thanks to all of you who made these two days possible. Starting with our uh partners Accenture and Ant International. Thank them for their support and uh and partnership. Um, thank you to our incredible speakers and moderators who came from around the world to share their time, insights, and visions for the future of AI. And to my fellow co-chairs, Andrew Nusca, Shakana, uh Jeremy Khan, thank you for your leadership and for guiding us through these bold, complex, and exciting conversations. And most of all, thank you, uh all of you, our audience, for showing up with curiosity and candor. We hope you're leaving with new connections, changed perspectives, and a sharper sense of what comes next. Uh, and speaking of what's next, please do save the date for our upcoming events. We'll gather for the Fortune Innovation Forum in Kuala Lumpur uh on November 17th and 18th, and we'll be in San Francisco for Brainstorm AI on December 8th and 9th. We hope to see you there or at another Fortune event very soon. Thanks very much everyone, and enjoy your evening.